Documentation for modified -new- QRF model

¹ Mount Hood Environmental, PO Box 1303, Challis, Idaho, 83226, USA
² Mount Hood Environmental, 39085 Pioneer Boulevard #100 Mezzanine, Sandy, Oregon, 97055, USA
³ Mount Hood Environmental, PO Box 4282, McCall, Idaho, 83638, USA

^✉ Correspondence: Bryce N. Oldemeyer <Bryce.Oldemeyer@mounthoodenvironmental.com>, Mark Roes <mark.roes@mthoodenvironmental.com>

1 Background

Habitat covariates recently used in six quantile random forest (QRF) capacity models (Chinook salmon and steelhead; summer parr, winter presmolts and redds) were chosen because of their high predictive power to estimate capacity across the Columbia River Basin (cite IRA?). However, a subset of the covariates included the QRF models were not necessarily useful for restoration project monitoring, or to describe target conditions for restoration design due to the habitat covariate not being easily manipulated by project actions. Additionally, some of the CHaMP covariates used in previous models are difficult to replicate or measure using streamlined fish habitat protocols (DASH - Carmichael et al. 2019). To increase the utility of the QRF model for project monitoring/design and future data collection efforts, we explored alternative covariates to include in the QRF models that: 1) maintained high predictive power, 2) were informative for restoration efforts and monitoring, 3) could be calculated from DASH surveys, 4) were not missing an overabundance of data, and 5) were not highly correlated with other covariates in the models. Using this criterion, we developed six modified QRF model that were more informative for restoration design and monitoring, included covariates that could be calculated using newly developed stream habitat protocols, and maintained a similar level of predictive power as the original QRF habitat capacity models.

Similarly, a random forest extrapolation model was used to predict capacity estimates across larger scales where CHaMP or DASH data are absent (cite IRA). We revisited the globally available attributes (GAA) included in the original random forest extrapolation model and made minor modifications so GAA’s included in the extrapolation model better aligned with the modified QRF model covariates.

Below is a brief outline documenting these efforts, as well as a comparison of extrapolation estimates for the original and modified QRF models for eight watersheds in the Upper Salmon River region.

2 Covariate selection

The QRF model was fit using a revised covariate selection process that placed more emphasis on compatibility with future data collection via DASH and the ability to predict restoration effects. Habitat data collected by CHaMP and other sources (e.g. NorWeST stream temperature) were subset to a list of potential covariates that could be reproduced using DASH data collection. This provides opportunity to collect new paired fish and habitat data using DASH protocols, reducing the reliance of the QRF model on the CHaMP habitat data.

Below is a rubric that was used to help inform the covariate selection process for each of the six models (Chinook and steelhead; winter, summer, and redds)

1. Strength between covariate and response variable (based on MIC score)

2. Informative for restoration efforts (Yes/No)

3. Could be calculated using DASH data (Yes/No)

4. How much data were missing and/or the amount of “0”s?

5. How correlated was the covariate with other covariates in the original QRF model, and covariates within the same model?

An oversimplified example of the theoretical covariate selection process might unfold as follows. In the original QRF model, discharge might have been included in the model because it had a high MIC score and it made biological sense. Unfortunately, discharge isn’t that informative for restoration efforts because most restoration efforts can’t create water. Discharge (like many habitat covariates) are highly correlated to other habitat covariates, but these other covariates maybe have been left out of the original QRF model for any number of reasons (highly correlated with other covariates already in the model, redundant, etc.). Using the rubric, we found that average thalweg depth had a MIC score that was nearly as high as discharge, it was informative for restoration, it could be calculated with DASH, and the two covariates were highly correlated (the high correlation is likely why average thalweg depth was left out of the original QRF model). Based on all the information above, we would substitute mean thalweg depth for discharge in that particular model. Repeat this process for all other QRF covariates for each of the six models.

Briefly discuss how we compared outputs between species and settled on a joint model.

3 QRF Model fit

Talk about relative importance and pdp plots.

4 Extrapolation model

The spatial extent of QRF capacity predictions was limited to reaches CHaMP habitat data, so capacity for all wadable streams in the Columbia basin was estimated through the development of an extrapolation model. This model used ‘globally available attributes’ (GAAs) obtained from a stream layer created by Morgan Bond and Tyler Nodine based on the National Hydrography Dataset High Resolution 1:24,000 line network to estimate capacities predicted by the QRF model at the 200 meter reach scale.

4.1 Extrapolation comparison

Figure 4.1: Extrapolations of habitat capacity for Chinook salmon, by life-stage, for the eight watersheds within the Upper Salmon River Basin using the modified models.

Figure 4.2: Extrapolations of habitat capacity for steelhead, by life-stage, for the eight watersheds within the Upper Salmon River Basin using the modified models.

5 Habitat Capacity Estimates

5.0.1 Chinook Salmon

Table 5.1: Predicted Chinook salmon habitat capacity by life-stage and watershed using the modified models.

Watershed	Juv summer capacity	Summer SE	Juv winter capacity	Winter SE	Redd capacity	Redd SE
EF Salmon	1,926,623	226,925.7	138,214	32,880	402	21
Lemhi	786,452	62,659.8	141,515	15,359	353	11
NF Salmon	339,275	50,147.9	70,462	10,409	166	8
Pahsimeroi	265,099	18,409.2	86,999	9,781	139	4
Panther Cr	1,219,542	118,369.5	201,265	22,296	448	17
Upper Salmon	3,301,286	352,419.5	166,522	45,582	575	29
Valley Cr	1,902,198	207,362.9	115,517	32,535	394	20
Yankee Fork	2,144,056	274,555.8	119,298	28,783	438	23

5.0.2 Steelhead

Table 5.2: Predicted steelhead habitat capacity by life-stage and watershed using the modified models.

Watershed	Juv summer capacity	Summer SE	Juv winter capacity	Winter SE	Redd capacity	Redd SE
EF Salmon	252,597	15,520.5	337,682	36,795	413	24
Lemhi	310,577	9,082.3	363,898	27,441	441	18
NF Salmon	242,471	18,381.8	313,118	27,955	323	22
Pahsimeroi	159,705	6,225.1	205,921	13,951	198	8
Panther Cr	268,476	13,598.0	339,671	19,946	317	15
Upper Salmon	243,548	14,843.6	310,879	39,013	452	32
Valley Cr	176,048	10,707.6	288,579	31,329	365	26
Yankee Fork	197,926	12,378.9	341,310	38,555	449	36

5.1 Comparison with previous extrapolation

Below are comparisons with the results from the previous QRF model and random forest extrapolation.

5.1.1 Chinook

Figure 5.1: Comparison of Chinook salmon habitat capacity estimates between revised and original model extrapolation, by life-stage, for the eight watersheds within the Upper Salmon River Basin.

Table 5.3: Estimated chinook capacities and comparison with previous random forest extrapolations for eight watersheds

Model	Watershed	Predicted capacity	Capacity % change	Predicted capacity SE	SE % change
Juv summer	EF Salmon	1,926,623.4	112	226,926	186
Juv summer	Lemhi	786,451.7	112	62,660	172
Juv summer	NF Salmon	339,275.4	13	50,148	100
Juv summer	Pahsimeroi	265,099.3	45	18,409	54
Juv summer	Panther Cr	1,219,541.6	21	118,369	33
Juv summer	Upper Salmon	3,301,286.0	163	352,419	205
Juv summer	Valley Cr	1,902,197.5	152	207,363	191
Juv summer	Yankee Fork	2,144,056.4	222	274,556	284
Juv winter	EF Salmon	138,214.5	0	32,880	139
Juv winter	Lemhi	141,514.7	-8	15,359	127
Juv winter	NF Salmon	70,462.3	28	10,409	106
Juv winter	Pahsimeroi	86,999.4	-8	9,781	44
Juv winter	Panther Cr	201,265.5	29	22,296	122
Juv winter	Upper Salmon	166,521.7	-29	45,582	87
Juv winter	Valley Cr	115,516.8	-12	32,535	145
Juv winter	Yankee Fork	119,298.3	20	28,783	122
Redds	EF Salmon	401.9	-13	21	-29
Redds	Lemhi	353.0	5	11	19
Redds	NF Salmon	165.7	-5	8	-4
Redds	Pahsimeroi	139.4	25	4	17
Redds	Panther Cr	447.8	-4	17	-14
Redds	Upper Salmon	575.0	-20	29	-41
Redds	Valley Cr	393.7	-29	20	-44
Redds	Yankee Fork	438.2	-38	23	-59

5.1.2 Steelhead

Figure 5.2: Comparison of steelhead habitat capacity estimates between modified and original models extrapolation, by life-stage, for the eight watersheds within the Upper Salmon River Basin.

Table 5.4: Estimated steelhead capacities and comparison with previous random forest extrapolations for eight watersheds

Model	Watershed	Predicted capacity	Capacity % change	Predicted capacity SE	SE % change
Juv summer	EF Salmon	252,597.2	-31	15,521	-1
Juv summer	Lemhi	310,577.2	-15	9,082	11
Juv summer	NF Salmon	242,471.4	-5	18,382	34
Juv summer	Pahsimeroi	159,705.1	-18	6,225	14
Juv summer	Panther Cr	268,475.9	-8	13,598	42
Juv summer	Upper Salmon	243,548.0	-31	14,844	-11
Juv summer	Valley Cr	176,047.8	-28	10,708	-11
Juv summer	Yankee Fork	197,926.3	-29	12,379	38
Juv winter	EF Salmon	337,681.9	-14	36,795	30
Juv winter	Lemhi	363,897.6	-8	27,441	52
Juv winter	NF Salmon	313,118.1	-1	27,955	4
Juv winter	Pahsimeroi	205,921.2	-4	13,951	21
Juv winter	Panther Cr	339,671.3	8	19,946	25
Juv winter	Upper Salmon	310,878.6	-26	39,013	20
Juv winter	Valley Cr	288,579.2	-14	31,329	10
Juv winter	Yankee Fork	341,309.6	-18	38,555	-7
Redds	EF Salmon	413.2	-13	24	-13
Redds	Lemhi	441.4	10	18	12
Redds	NF Salmon	323.4	-10	22	22
Redds	Pahsimeroi	198.5	2	8	-8
Redds	Panther Cr	317.0	-7	15	15
Redds	Upper Salmon	452.1	-11	32	-6
Redds	Valley Cr	365.0	-20	26	-7
Redds	Yankee Fork	448.9	-25	36	13

6 Supplemental figures and tables

6.1 Capacity by stream

6.1.1 Chinook

6.1.2 Steelhead